Intel x86 and IBM Power CPUs: Which, when, why?
There are only three processors left in the market for mission-critical applications. Which should you chose ?
There are only three processors left in the market for mission-critical applications. The Intel x86 processor dominates mid-range and small servers, desktops, laptops and notebooks. The IBM mainframe zEC12 processor still dominates extreme workloads for data processing. And the IBM Power processor, which has taken over Unix workloads previously dominated by Sun, HP and other fallen Unix leaders.
For the general market, small business and departmental applications, the choice boils down to Intel x86 or Power. The selection criteria between these two technologies lie in their heritage. Unfortunately too many workloads are being deployed inappropriately, simply based on market momentum or perceived cost advantages. If we just look at the heritage of the two technologies we can get a perspective on which choice provides the most cost-effective performance for given workloads.
Intel does everything well
Intel x86 made its mass market debut in 1982 in the IBM PC, known as the Intel 8088. That’s over 30 years’ ago. Since then, Intel has made significant enhancements to the instruction set architecture to keep it relevant in a market where 18 months is obsolescence. They moved from 16-bit to 32-bit and now 64-bit. They have added hardware accelerators on the chip, and they have even built Risc instruction into the architecture to speed up some of the Cics instructions that plagued performance. Intel x86 runs in everything from supercomputers to servers to desktops and laptops. Intel does everything well. Some would say “good enough”.
Download Terry Keene's e-book
- Navigating your Future within an SOA information manufacturing system
Power favours data and high performance computing
The current IBM Power architecture design began in 1997 and the processor was announced in 2001 as the Power4. It was the first multicore processor in the industry. The design team was recruited from IBM NY labs in Poughkeepsie, Kingston, Fishkill, and from the IBM AS/400 lab in Rochester, MN. This team included the best in mainframe design, supercomputing design and chip fabrication design. From its origin, the Power architecture was designed by this team to do two things exceptionally well; data and high-performance computing.
Download resources on virtualisation
Comparing the x86 and Power processors on a micro-benchmark level will show little raw performance advantages for either. Comparing the two using enterprise workloads will demonstrate a significant advantage for Power in data workloads such as databases, data warehouses, data transaction processing, data encryption/compression, and certainly in high-performance computing, which most in business think of as analytics.
Virtualisation makes the real difference
In the database workloads the real difference is virtualisation. Intel x86 virtualisation consists of many third-party add-ons like VMware, KVM, OVM, XEN, Hyper-V and the like. All of these constitute hypervisors, mainly Linux OS kernels with the exception of Hyper-V from Microsoft, customised to act as an agent for virtual machines interacting with the hardware and system I/O. This type of virtualisation certainly improves the efficiency of x86 platforms over a single server with a single OS and a single application, but the overhead of the software hypervisor limits the utilisation and scaling of the host server. Generally the average resource utilisation climbs to 20-25% versus the average 5% -10% utilisation of a single server. The hypervisor can consume as much as 25% of the system resources when acting as the VM agent.
Hypervisors
The real secret sauce of the Power architecture is the Power hypervisor. The Power design team built support for the Power hypervisor into the chip from the beginning. Unlike Intel hypervisors, the Power hypervisor runs as a small piece of firmware code interacting with the hardware and guest VMs running on the platform. The result is minimal overhead, at most 2%, with maximum efficiency for CPU resources, memory and virtual I/O. And because the Power hypervisor was designed in conjunction with the system architecture, it is always running, even for a single OS instance, regardless of the OS – be it AIX, Linux or IBM i.
The selection between Intel x86 and Power for workload deployment depends on the workload profile, architecture and OS support, and scalability requirements. If the application is written exclusively for Windows, choose Intel. But deploy it on VMware or similar x86 hypervisor. If the application is data transaction processing with a database or data warehouse, or data analytics, then Power is by far the better platform. It’s horses for courses.
Terry Keene (pictured) is CEO of iSys (Integration Systems, LLC)
